Isospectrality and heat content M. van den Berg, E. B. Dryden * T. Kappeler School of Mathematics, University of Bristol ^ ! University Walk, Bristol BS8 1TW, UK ^ M.vandenBerg@bris.ac.uk u < in 57 Ctf o X t Department of Mathematics, Bucknell University Lewisburg, PA 17837, USA emily . dryden@bucknell . edu Institut fiir Mathematik, Universitat Zurich Winterthurerstrasse 190, CH-8057 Zurich, Switzerland C/} \ thomas.kappeler@math.uzh.ch r4 15 April 2013 Abstract We present examples of isospectral operators that do not have the same heat content. Several of these examples are planar polygons that are •-— ' . isospectral for the Laplace operator with Dirichlet boundary conditions ^1 ' These include examples with infinitely many components. Other pla- • , nar examples have mixed Dirichlet and Neumann boundary conditions. We also consider Schrodinger operators acting in L 2 [0, 1] with Dirichlet ^— * ' boundary conditions, and show that an abundance of isospectral defor- mations do not preserve the heat content. Mathematics Subject Classification (2010): 58J53; 58J50; 35P10. Keywords: Isospectral, heat content, isoheat, polygons, Schrodinger operators. * Emily Dryden was partially supported by a grant from the Simons Foundation (210445 to Emily B. Dryden). She also thanks the School of Mathematics at Trinity College Dublin for their hospitality. 1 Partially supported by the Swiss National Science Foundation. 1 Introduction Let (M, g) be a compact m-dimensional Riemannian manifold, not necessarily connected and with a topological boundary dM. Let —Am be the associated Dirichlet Laplace-Beltrami operator acting in L 2 (M, dx), where dx is the volume measure on M induced by g. The spectrum of —Am is discrete and we denote the eigenvalues by Ai(M) < A2(M) < A 3 (M) < ... , where each eigenvalue is repeated according to its multiplicity. Two Riemannian manifolds Mi and Mi are isospectral if Xj(M\) — Aj(-Ma), j = 1,2,.... We refer to [TU] and the references therein for a survey and details on isospectrality. We denote the heat trace for M by oo Z M (t) = Tr L2 (e 4A «) = J2 e" tAj(M) , t > 0. i=i The heat trace for M is a spectral invariant, and we note that Zm(£) determines the spectrum, i.e., the eigenvalues of —Am together with their multiplicities. Moreover its asymptotic behaviour as t J, contains geometric information on M. It has been shown that if dM is smooth then there exists an asymptotic series such that for any I E N i ZAt{t) = ^2 a i (M)t {l - m)/2 + 0(t (/+1 - m)/2 ), 1 1 0, (1) 2=0 where the a,i(M) are locally computable spectral invariants. See [8] for further details. Of much interest is the heat content of M. Let u : M X [0, oo) — > R be the unique solution of 3u A M u = — , t > 0, (2) with initial condition limu(-;i) = 1 ini 2 (Af), (3) and Dirichlet boundary conditions u(x;t) =0, i£ dM, t> 0. (4) The heat content QM(i),t > 0, is defined by QM(t) = / u(x;t)dx Jm (see [HE] and the references therein). It has been shown that if dM is smooth then there exists an asymptotic series such that for any JeN Qm{1) - ]T b t (M)f/ 2 + 0(t< /+1 )/ 2 ), t i 0, (5) i=0 where the bi(M) are locally computable heat content invariants. We refer to [2 for the existence and to pQ for the calculation of the first few coefficients. The solution u of (0) , (H]) and Q can be expressed in terms of the spectral resolution {<f>j t M, Aj(M)} 0I -Am, where {</>i,m, 4>2,m, </>3,m, • ■ • } denotes an or- thonormal basis of eigenfunctions corresponding to Ai (M) < A2 (M) < A3 (M ) < .... Let pM{x,y;t),x £ M,y £ M,t > denote the Dirichlet heat kernel for M. Then OO p M (x,y;t) =J2 e ~ tXj(M) h,MW c f > jMy)- 3=1 We have that u(x;t) = PM {x,y;t)dy = Ve' tt ' (M) ( / faMvWv) <t>j, M (x). Jm j=1 \Jm J By definition of the heat content, 00 , . -.2 QM(t) = £ e~ tXAM) I I <P ] ,M(x)dx) . (6) We say that Mi and Mi are isoheat if for all t > 0, Qm x (t) = QM 2 (t)- The number § M <t>j,M{y)dy is just the j th Fourier coefficient of the initial datum, which is the constant function 1 on M. It follows from © that if all eigenvalues are simple, and if this Fourier coefficient is nonzero for all j £ N, then Qm (t) determines the spectrum. Note that the invariants one can read off from (j6j) are not local. From formula ([5]) one cannot decide if Quit) is a spectral invariant like Zm{£). Nevertheless it shares some of its features, like the existence of an asymptotic expansion for t ], 0. The aim of this paper is to explore the relation between isospectrality and the property of being isoheat. It is known that isoheat does not imply isospectral [9]. Below we give two new examples demonstrating this, but first we make a general observation. If M does not have a boundary, we have that \\(M) = with <p\ t M — |M| -1 / 2 up to a sign, where \M\ = J M ldx. All but the first Fourier coefficients equal 0, and so Q M (t) = \M\. Proposition 1. Let Mi and Mi be manifolds without boundary and with the same volume. Then Mi and Mi are isoheat. Example 2. Let Mi and Mi be compact Riemannian manifolds without bound- ary such that \Mi\ = \Mi\, Mi is connected, and Mi is not connected. Mi and Mi are not isospectral since Mi has a first eigenvalue with multiplicity 1 whereas Mi has a first eigenvalue with multiplicity at least 2. By Proposi- tion [7J Mi and Mi are isoheat. If M3 is a compact Riemannian manifold with smooth boundary then Mi x M3 and Mi x M3 are isoheat manifolds with bound- ary which are not isospectral. Also note that if Mi and Mi are isoheat and M3 and M4, are isoheat then Mi x M3 and Mi x M4 are isoheat. Our second example consists of planar polygons. Both the heat trace and heat content are well defined and admit expansions similar to ([T]) and ([5]) for these planar domains. Recall that Neumann boundary conditions are given by 3u — (x;t) =0, x£dM, t>0, (7) on where n is an outward unit normal vector field on dM . Example 3. Let A be a rectangle with edges of lengths 1 and 2 and with Dirich- let boundary conditions. Let B be the disjoint union of two squares of edge length 1, where each square has Dirichlet boundary conditions on three edges and Neumann boundary conditions on the remaining edge. The heat flow in A is symmetric with respect to the axes of symmetry, so in particular satisfies Neumann boundary conditions along the short axis. Thus the heat content of A is the same as the heat content of B. We observe that \i(A) = 57r 2 /4 has multiplicity 1 and Ai(-B) = 57r 2 /4 has multiplicity 2. Hence A and B are not iso spectral. It was shown in [9 that a Sunada construction involving finite coverings will not produce isospectral manifolds that are not isoheat. However, the general question of whether there exist isospectral manifolds with different heat contents was hitherto unanswered. Note that the question has been resolved in [14] for weighted graphs: McDonald and Meyers construct weighted graphs associated to the principal example in [5J, which is a simplified version of the drums of |11| . These graphs are planar, isospectral, and non- isometric; using the heat operator naturally associated to such a weighted graph, the authors calculate the heat content of both graphs. They show that the fifth coefficient 64 in the small-time asymptotic expansion of the heat content differs for the two graphs, hence they are not isoheat. In Section [2J we present multiple examples to show that isospectral does not imply isoheat in the manifold setting. Our examples consist of planar domains, and mostly of planar polygons. Some have all Dirichlet boundary conditions and others have mixed boundary conditions. We present examples in which both sets of the isospectral pair consist of the same number of connected components, and an example in which one set is connected and the other is disconnected. To show that our sets are not isoheat, we use two approaches: (i) we either show that the coefficients in the small-time asymptotic expansion of the heat content differ for the two sets, or (ii) we consider the large-time behaviour of the heat content. In particular, for the example consisting of disconnected sets with infinitely many components in each, we use the large-time behaviour. The construction of these infinite disconnected sets involves a family of self-similar polygons, and it is interesting to obtain the first few terms in the expansion for t \. of the heat trace and heat content of such a "fractal" family. In Section |31 we do so for the heat trace. In Section @] we consider Schrodinger operators L q — — 4-^ + q acting in L 2 [0,1] with Dirichlet boundary conditions at and at 1. We assume that q € £ 2 [0, 1]. L q has discrete spectrum and its eigenvalues are simple: Ai(g) < A 2 (<?) < A 3 (g) < . . . (e.g., [HI Chap. 2]). As in the case with —Am one defines the heat trace Z q (i) and the heat content Q q (t) of L q . Two potentials q\ and (72 in L 2 [0, 1] are isospectral if Xj{qi) = Aj(g2) for all j e N. We say that q\ and g 2 in L 2 [0, 1] are isoheat if for all t > 0, Q qi (t) = Q q2 (t). We wish to explore the relationship between isospectrality and the property of being isoheat using the fact that these model operators have simple eigenval- ues. First we note that as in the case of — Am, the heat content of L q has a representation of the form Q q {t)^Y,^ tx,{q) {sy^ x ) dx ) where for any j G N, 4>j_ q denotes the eigenfunction corresponding to Xj(q) normalized by J Q (f)j, q (x) 2 dx — 1 and </>j i9 (0) > 0. Note that 4>j^ q € H 2 [0, 1] and hence by the Sobolev embedding theorem, cj)j tq e C^O, 1]. According to [8] for any smooth potential q, Q q (t) admits an asymptotic expansion of the form Q q (t) = Y / b M tl/2 + o(t {I+1)/2 ), U0, (8) i=0 where / e N and b (q) — l,bi(q) = — 47r~ 1 / 2 and 62(9) = — J q(x)dx. In particular it follows that the mean of q is an invariant of the heat content. Since b\(q) ^0 we have that the set ,] q = {j e N : J 4>j. q 7^ 0} is infinite. We remark that from the asymptotics \j(q) = ti 2 ] 2 — ^2(9) + °(l)jj — > 00 [TH) p. 35], it follows that the mean is also a spectral invariant (see also [5]). There is a natural pair of isospectral isoheat potentials. Consider for any q € L 2 [0,1] the reflected potential g* (x) = q(l — x),0 < X < 1. According to [T5J p. 52], g* and q are isospectral. Furthermore, the eigenfunctions corresponding to g*, denoted <fij tq%l (x), are related to the eigenfunctions corresponding to q by <f>j,q*(x) = (—iy +1 <j)j tq (x) [H3 p. 42]. As a consequence f Q 4>j,q,{ x )dx = (—iy +1 f 4>j : q{x)dx, implying that g and g* are also isoheat. In Section |H we give an abundance of isospectral deformations that are not isoheat. 2 Heat content and planar domains As mentioned in the Introduction, both the heat trace and the heat content are well defined and admit expansions similar to ([1} and ([5]) in the case of planar polygons with associated Dirichlet Laplace operator. These are bounded regions in R 2 , not necessarily connected nor simply connected, such that each compo- nent of the boundary consists of vertices that are connected by straight line seg- ments. It was shown in |3] that if P is a polygon in R 2 with vertices Vi , V2, . . . , V n with corresponding inward pointing angles 71 , 72 , . . . , J n then there exist con- stants di(P),d2(P) depending on P such that 7 (+) l P l 1 \ dP \ y* -7. P[ ' 47rf 8( 7 ri) 1 /2 2-. 247r 7l < d 1 {P)e- d2(p ^ t , £>0, (9) where \P\ is the area of P and \dP\ is the sum of the lengths of the components of the boundary. For example if dP is connected then \dP\ =J27=i Wi~ K+i| + \V n -Vi\. For the heat content it was shown in [I] that there exist constants ds(P), d±(P) depending on P such that 2/1/2 n Mt) -\P\ + -^j2\9P\-tJ2<7i) i—1 < d 3 (P)e- d ^ p ^ t , t > 0, (10) where c( 7 ) = r . 4sinh(( "~ 7) " } dx. Jo (sinh(7ra;))(cosh(7a;)) It follows from ([TUf that we have an expansion of the form (|S|), where &o(P) = \P\, 6i(P) = -2 7 r- 1 /2| 5P | j &2 ( P ) = S™ =lC ( 7i ), 6s(P) = 6 4 (P) = ••• = 0. Both the proofs of © and ([TU)l rely on detailed calculations using the repre- sentation of the heat kernel for the infinite wedge as a Kontorovich-Lebedev transform. It was reported in [TS] that D. B. Ray obtained the contribution of the vertices Yn=i 247r 7i m © using this transform. We now give several examples of isospectral planar domains that are not isoheat. In the first example, one domain is connected and has only Dirichlet boundary conditions, but its isospectral partner is disconnected and has mixed boundary conditions. Example 4. Let A be a rectangle with edges of lengths 1 and 2 and with Dirichlet boundary conditions on all edges. Let B be the disjoint union of two unit squares, one with all Dirichlet boundary and one with Dirichlet boundary conditions on three edges and Neumann boundary conditions on the remaining edge. The even eigen] 'unctions in A satisfy Neumann boundary conditions along the short axis of symmetry. There is a one-to-one correspondence between these even eigenf unctions in A and eigen] 'unctions of the square in B with mixed boundary conditions. A similar situation holds for the odd eigenf unctions in A and the eigenf unctions of the square in B with Dirichlet boundary conditions. Thus A and B are isospectral. However, note that \dA\ = 6. Hence by (|10|) . the coefficient oft 1 ' 2 in the expansion for Q a (t) is — 127T -1 ' 2 . The square in B with Dirichlet boundary conditions has perimeter 4, and so this gives a contribution — 87T -1 ' 2 to the coefficient oft 1 ' 2 . The heat content of the remaining component in B can be computed explicitly by separation of variables. In Section^ below we show that this results in a contribution —Qn^ 1 ' 2 to the coefficient oft 1 ' 2 . Hence the coefficients of t 1 ' 2 are — \2ir~ 1 ' 2 and — 147T -1 ' 2 for A and B respectively. So A and B are not isoheat. The domains in the next example both consist of one component and have mixed boundary conditions. Example 5. The half-disks with mixed boundary conditions as shown in Figure [7] are isospectral 1131. The boundary of A is endowed with Dirichlet boundary conditions exactly where the boundary of B is endowed with Neumann boundary conditions, and vice versa. To see that A and B are not isoheat we consider the difference QA(t) — Qsit). To this end we place vertices on the boundaries of the half-disks at the points where the equator meets the circle ( "endpoint vertices") and anywhere else the boundary conditions change. We see that for each possible pairing of boundary conditions at a vertex, there is a bijection between the non- endpoint vertices in A and the non- endpoint vertices in B. Moreover, each of A and B have one endpoint vertex where a Dirichlet and a Neumann edge meet. This leaves one endpoint vertex Va in A where two Neumann edges meet, and one endpoint vertex Vb in B where two Dirichlet edges meet. Since Va has a boundary neighbourhood with Neumann boundary conditions there is no contribution to 62 {A) from this vertex. On the other hand the geometry of the boundary near Vb is in first approximation that of a wedge with angle ir/2. This suggests that an angle contribution c{ir/2)t = 4t/ir will show up in Qs(i), and that Qa(£) — Qsit) = —At/ir + o(t) so that A and B are not isoheat. This can be made rigorous using the Brownian motion tools from Q Ef. In the remaining examples, we take a different approach to that taken above to show that these isospectral sets are not isoheat. Instead of using the coeffi- cients in the small-time asymptotic expansion of the heat content, we consider Figure 1: Isospectral half-disks A and B with mixed boundary conditions, where solid lines represent Dirichlet boundary conditions and dotted lines represent Neumann boundary conditions [TJ] the behaviour for t —¥ oo. The heat content function behaves like its leading term e~ tXl (J 4>i) 2 , so we show that {J 4>i) 2 differs for our isospectral domains. The reader can easily check that the coefficients in the small-time expansion actually match for the isospectral sets considered below. Chapman !7j constructed several examples of isospectral, non-isometric drums. We call a disconnected drum a band. We denote the two-piece band consisting of the square of edge length 1 and the right isosceles triangle with area 2 by A, and the pair consisting of the rectangle of edge lengths f and 2 and the right isosceles triangle with area 1 by B. Figure 2: Chapman's isospectral two-piece bands A and B Theorem 6. The two-piece Chapman bands A and B are isospectral but not isoheat. In particular, we have the following expressions for their heat contents. (i) For t — > 00, (ii) For t — > oo, l(*) 1024 -5tt 2 £/4 0(e -2^t\ „ , s 1152 Qs(t) = -^e 5 " 2t / 4 + 0(e~ 2 * 2t ). (11) (12) Proof. The eigenvalues for the rectangle with vertices (0, 0), (a, 0), (0, 6), (a, b) are 7r 2 ((^) 2 + ( : y) 2 ), where n € N and m e N. The corresponding eigenfunctions are <f> n ,m(x,y) = sin( 2 ^ £ ) sin( I!: ^). For a right-angled isosceles triangle with vertices (0,0), (c, 0), and (c,c), the eigenvalues are 7r 2 ((^) 2 + (^) 2 ), where ieN and j £ N with i > j. The corresponding eigenfunctions are ipij(x,y) — sin(^) sin(^) - sin(^) sin(^). We find that the first few eigenvalues of A and B are 57r 2 /4, 2tt 2 , 5tt 2 /2, f 3tt 2 /4, f 7tt 2 /4, .... Note that 5?r 2 /4 is the lowest eigenvalue of the triangle in A and of the rectangle in B. We find that for the rectangle \\<f>n,m\\ 2 2 = ab/4, and ab l <t> n , m {x,y)dxdy = -^(1 + i-iy 1 - 1 )^ + (-I)™ -1 ). [0,o]x[0,6] mn7T2 It follows that with a = 1, b = 2, m = n = 1, and recalling that </>i,b is normal- ized in L 2 (B), 2 128 1152 '1,B — A ,, B I 7T 97T 4 It is convenient to consider the general case where \\{M) has multiplicity 1. By Bessel's inequality < e -tA 2 (M)| M | IM We conclude that if Ai(-M) has multiplicity 1 then Q M {t)=e- tx ^ M Uj <p 1M {x)dx\ +0{e- tx ^). (13) This proves (ii) since \\{B) = 57r 2 /4, and A 2 (5) = 27r 2 . We find for the right-angled isosceles triangle with edges of lengths c, c and cy/2 respectively that UlMS =c 2 / <b / <%((sin M sin( J7ry )) 2 + (sinOV.) sin(^)) 2 J [0,1] J[0,x] c 2 — 2 sin(i7ra;) sin(j7ry) sin(j7T2;) sin(i7ry)) = — , (14) and (15) dx / dyip 2 ,i{x,y) [0,c] J[0,x] = c / dx / <iy(sin(27ra;) sin(7rj/) — sin(27rj/) sin(7rx)) «/[0,l] J[0,a:] _8c^ 3tt 2 ' It follows by HUD and ([T5J for c = 2 that \ 2 1024 This proves (i) by ([13) since Ai(A) = 5tt 2 /4, and A 2 (A) = 2tt 2 . D Using the fact that A and B are isospectral one can construct a whole family of isospectral bands as follows. Add to A and B a disjoint square with area 1/2, and now use the fact that this square with the right angled isosceles triangle in set B is isospectral to 2~ 1//2 B, where 2~ 1 / 2 B means that we scale the original polygons in B by a factor of 2~ 1 / 2 . We proceed by induction to obtain two bands C and D which are isospectral and not isoheat. C is a disjoint union of squares with areas 2~ J : j € {0, 1,2,...} together with a right angled isosceles triangle of area 2. D is a disjoint union of rectangles R\, i?2> • • • > where i?j has area 2 1-J and diameter 2~^ 2 ^/Z. Corollary 7. TTie infinite Chapman bands C and D are isospectral and not isoheat. In particular, we have the following expressions for their heat contents. (i) For t — > oo, (ii) For Moo, Qc(t) = ^e- 5 ^/ 4 + 0(e- 2 ^). Q D (t) = ^e-^ + 0(e-^). Proof. Note that 5-7r 2 /4 is the lowest eigenvalue of the triangle in C and of the largest rectangle in D. So (f c cjy lfi ) = (/ A 0i,a) = ^pr and (J D (j) 1:D ) = (J b <I>i,b) — if^r- Since \C\ = \D\ < oo we may apply (1T3"]) to complete the proof. D Given the "fractal" nature of C and D it is of independent interest to obtain the first few terms in the expansion for t I of the heat trace Zc (t) = Zu (t) as well as the heat contents Qc(t) and Qo{t). We illustrate the techniques involved for the heat trace only. See Theorem |H] and its Corollary [TU] in Section 121 The original drums of Gordon et al. |11) . the various examples of planar isospectral drums subsequently constructed by Buscr ct al. 6 , and Chapman's modifications can all be viewed through the lens of graph theory. Herbrich [12] works in the context of mixed boundary conditions, showing how to translate the transplantability condition satisfied by the examples in [HI [7J [IT] into graph theory. This led to an algorithm for finding transplantable pairs, a classification in terms of induced representations, and ways of generating new pairs from known ones. In particular, Herbrich finds twelve versions of the drums of Gordon et al., of which ten have mixed boundary conditions (see Fig. 5.9 in [H]). One of these bands was already found by Parzanchevski and Band |17,. Imitating Chapman's modification of the original drums of Gordon et al., we obtain the isospectral three-piece bands with mixed boundary conditions shown in Figure 131 Corollary 8. Let A be the disjoint union of a square with area I, a right-angled isosceles triangle with area 2 and a right-angled isosceles triangle with area 1. One edge of the square has Neumann boundary conditions, and the V2 edge in the right-angled isosceles triangle also has Neumann boundary conditions. Let B denote the disjoint union of a right isosceles triangle with area 1, the rectangle of edge lengths 1 and 2, and the right isosceles triangle with area ^ One V2 edge in the right-angled isosceles triangle of area 1 has Neumann boundary conditions, and one edge in the right-angled isosceles triangle with area \ also has Neumann boundary conditions. All other edges in both A and B have Dirichlet boundary conditions. Then A and B are not isoheat. In particular, we have the following expressions for their heat contents. (b)£ Figure 3: Isospectral three-piece bands with mixed boundary conditions (i) For t —> co, (ii) For t — > oo. )x(«) = ^e^ + 0( e -n (16) Qb(*) 1664 -57T 2 t/4 0(e" (17) Proof. The large isosceles triangle in A has first eigenvalue 57r 2 /4. By doubling the square across the Neumann edge and recalling that the first Dirichlet eigen- function on that rectangle satisfies Neumann boundary conditions on the short axis of symmetry we have that this square also has a first eigenvalue 57r 2 /4. Furthermore, since the heat flow in the double of the square is also symmetric with respect to the short axis, we have by (fT"2|) that this square contributes I ' i^r e ~ 57r2 * /4 + 0(e~ 27r2 *) to the heat content in A. The contribution of the large isosceles triangle is given in (|11D . Addition of these contributions yields ([i"6| . Arguing in a similar fashion we see that both the rectangle and the large isosceles triangle in B have a first eigenvalue 57r 2 /4. The contribution of the isosceles triangle is | • ^e" 5 " */ 4 4- 0(e _27r *), whereas the contribution from the rectangle is given in (TT^I) . Addition of these contributions yields (fTT)) . □ These examples show that isospectral does not imply isoheat in the setting of planar polygons. It would be interesting to have an isospectral non-isometric pair of planar polygons with the same heat content. 3 Heat trace in the fractal setting Below we shall compute the first few terms of the asymptotic expansion for the heat trace of the band C as constructed in Section [2 Since C is isospectral to D, and since D consists of disjoint rectangles only, it is easier to consider the latter. Zoit) can be written as a triple sum over N 3 . However, it is easier to use the approach via the renewal equation [T|j] . This also allows us to consider general disjoint unions of self-similar polygons. Our set up is the following. 10 Let P be a polygon with area \P\ and boundary length \dP\. We abbreviate n 2 2 1=1 and 247r 7i ' so that © takes the form \R(t)\ <d l {P)e- d2{p)/t , t>0. (18) Let < a < 1 and denote by aPthe rescaled polygon {ax : x £ P}. We let P a be a disjoint union of sets a?P : j G N U {0}. It is easily seen that |P a | = (1 - o 2 )- 1 !^, \dP a \ = (1 - a)- x |ap|. Our main result is the following asymptotic expansion for the heat trace of P a . Theorem 9. There exists a 21og(l/a) periodic function ir at p such that for no, Zp.it) = ^ " ^f 2 +c Ql plogt + 7r a , P (togt) +0(e-*^/t«"*)), (19) where d 2 {P) is as in ^ and c a . P = ^-. (20) 2 log a Corollary 10. Lef C and D 6e t/ie bands constructed in Section Ql Then for where P is a rectangle with edges of lengths 1 and 2. Proof o/ Corollary \M Note that |P| = 2, \dP\ = 6, V(P) = 1/4, a = I/a/2, and D = P 1 /^2- Formula (j2"Tj) follows from Theorem |H] and the fact that C and D are isospectral. □ Proof of Theorem^ By scaling we have that for a > 0, t > 0, Z aP (t) = Z P (i/a 2 ). Since the heat trace is additive on a union of disjoint sets we have that Z Pa {t) = Z aPa (t) + Z P {t) = Z Pa {t/a 2 ) + Z P (t). (22) If we substitute 11 for the relevant expressions in (f2"2"j) then we obtain U{t) - U{t/a 2 ) = R(t). (23) Equation (|23[) is of renewal type, and has been studied extensively in relation to fractal geometry. See [TB] and the references therein. However, in order to invoke the Renewal Theorem [THl P- 198], two-sided decay estimates on R(t) are required. From (TT5)) we can deduce decay for £10, but no decay for t — > oo. Indeed, for £ — ► oo all of Zp(t),j^, and J t J /2 decay to 0. But the angle contribution V(P) is ^-independent and so the remainder does not decay for Moo. Iteration of (|2"3"j) yields for any j £ N, U{a 2 H) = C/(£/a 2 ) + ^ i?(a 2l £). i=0 Substitution of £ — a 2 8 yields 3 U{a 2]+2 6) = U{9) + Y, R(a 2l+2 6). i=0 We define j(t) e Z by 2i(t)+2 ^ . - 23(f) and 9{t) £ [l,a~ 2 ) by 0(4) = a- 2 ^*)- 2 i. Then for £ < 1, j(t) > and OO OO [/(£) = [/(#(£)) + ^ i?(a 2l+2 0(£)) - Yl R(a 2l+2 9{t)). i=0 i=j'(t)+l To complete the proof we have that for £ < 1 , CO CO CO J2 R{ a 2l+2 e(t)) < J2 \R{a 2l+2 e{t))\ = Y\R{^t)\ i=j(t)+l i=j(t)+l c <dx(P)X;e- d = 0(e OO ^-d 2 (P)/(a 2 *i) -d 2 (P)/(Q 2 t)x Finally we note that z i-> £?(e z ) is a 21og(l/a) periodic function. Hence £/(£) is 21og(l/a)-periodic in log£. This completes the proof. □ 4 Heat content and Schrodinger operators In this section, we return to the setting of Schrodinger operators L q = — -^s + <Z acting in L 2 [0, 1] with Dirichlet boundary conditions at and at 1. We assume that q £ L 2 [0, 1] and give an abundance of isospectral deformations that are not 12 isoheat. For this purpose consider the vector fields X n (q) — 2^(0„ !g ) 2 , n e N, introduced and studied in great detail in [IB]. Denote by s H> j n (s) the solution of Ts"tn{s) = X( ln (s)) with 7 „(0) = 0. By [B p. 91] where Note that and hence 7n(*)(») = ln{x;s) = -2d 2 x log 9 n (x;s), (24) 6 n (x; s) = 1 + (e s - 1) / 2(sin(7i7rr)) 2 dr. (25) l (x;s)= I 2(sm(nirr)) 2 dr + e s J 2(sm(nirr)) 2 dr, (26) min{l, e s } < 6 n (x; s) < max{l, e s }, s£l. (27) Furthermore an explicit computation shows that „ , , „ „ / sin(2n7ra;) \ 6 n (x;s =1+ e s -l )[l-x+— K - '-). 28 \ 2nn J In particular j n (x; s) is smooth in (x, s) E M x R. It has been shown in [18] that for any n G N and set, 7n(s) and the potential are isospectral. Hence 7 n (s) is an isospectral deformation. Theorem 11. For any n G N,7„(s) is not an isoheat deformation. The proof of Theorem 1111 relies on an explicit computation of the (Jo < f'j,'yn(s)(x)dx) near s = 0, based on the formulas for 7„(s) and Proof. Recall that (i) for the Schrodinger operator L q with q = 0, an orthonor- mal basis of eigenfunctions in L 2 [0, 1] is given by c/)j j q(x) = v2sin(j7r:z;), j > 1, and (ii) the isospectral deformation 7n(s) with initial condition 7 n (0) = is given by (JUJ) and (1231) . For j ^ n, the eigenfunction fij^ts) is computed in [15J p. 92] as . . r- , . sl . s v2sin(n7ra;) /" . , . . , . iPj^ji) s = V2sin(j7rx) - (e - 1) — — — / 2 sm(j7rr) sm(imr)dr, 8n{x;s) J x whereas for j — n, ^, 7 „ (s) (,). e ^-^ Sin( ^ ) »(»;s) These formulae are now used to expand the coefficients hj(s) = h n j(s) — Jo 4>j-in( s )( x )d x hi the heat content Q l7l ( s )if) = Ylj>i e_<J w ^j( s ) a ^ s = 0. Note that /i,(0) = V2 / sin(J7ra;)da: = — (1 - (-l) j ) Jo Jn 13 and 9 n (x; 0) = 1 whereas d s | s=Q e n (x;s) = f 2(sm{mrr)) 2 dr. J X Using that ds I g=o^".7n («)(*) = —pi sin(n7ra;) - \/2sin(n7ra)<9 s | s=0 &n(x; s) one gets fl I , f , V2 ! + (-!)" 1 5 - u nir 2 Similarly, for j ^ n one has <9 S | _ o 0j,7„(s)(a?) = — v2sin(n7ra) / 2sm(jnr)s'm(nirr)dr J X leading to d * \ s =o h o( s ) = -^ZZ S J^ n 1 1 y/2nn" where 5j^ n is the Kronecker delta. Hence for j ^ {2n,n}, hfc) whereas h 2 n (s) = 2n a 7r2 s 2 + 0(s 3 ) and 0(s : ), j even -^ + 0( s 2 ), j odd " ~^ + 0( S 2 ), nodd. n^TT* As a consequence Q ln ( s ){t) is not an isoheat deformation for s near 0. □ The expansions of the coefficients h n ^(s) at s = in the above proof show that for any j > 1 and n > 1 leading to the following. Corollary 12. For any n > 1 d s\ s=0 Q~f n ( S )(t) = o, voo. Formally, Corollary [T2l means that for any £ > 0, the differential doQ(t) of the map Q.(t) : q — > Q q (t) is not 1-1. In fact, the subspace of L 2 [0, 1], spanned by the elements X n (0), n > 1, is in the kernel of doQ(t). By setting up the maps Q.(t) in the appropriate spaces, these statements could be made precise. We briefly comment on a second family of vector fields, Y n (q),n > 1, intro- duced in [18, p. 108], Yn(q) = -2d x (a n - [a n ]</> 2 jq ) 11 where a n (x, q) = Vi{x, X„ {q),q)y 2 (x, X n (q),q) and [a n ] = L a n (x,q)dx. Here yi{x, A, q) and 2/2(2;, A, q) denote the fundamen- tal solutions of — y" + qy = Xy, satisfying the initial conditions yi(0, X,q) — 1, j/i(0, X,q) = 0, and 2/2(0, A, q) — 7 y' 2 {0,X,q) = 1 respectively. We see [TBI P- 65] that Y n (q) £ E for any q £ E and n > 1 where i?o denotes the subspace of even potentials, q(l — x) — q{x) for all a; € [0, 1], satisfying [q] = 0. Actually, by the construction of Y n (q), at any q £ Eq, the closure of the span of Y n (q), n > 1, is E 18, p. 107]. By 18, p. Ill], the initial value problem -f£(s)=^(£( S )), e(o) = o as has a unique solution £„(s) which exists for s € R satisfying (n - l) 2 7r 2 < n 2 7T 2 + s < (n + l) 2 7r 2 . The main feature of the flow £„(s) is that A„j(s) = Aj(£ n (s)) = j 2 7T 2 + sS n j where S n j denotes the Kronecker delta - see [T51 P- 108]. As for the eigenfunctions, recall that the eigenfunctions corresponding to <7* = q(l — x), denoted 4>j, q , (a;), are related to the eigenfunctions corresponding to q by (j>j, qt (x) = (—iy +1 cj)j yq (x) [JBJ p. 42]. As a consequence, J Q 4>j^ qt (x)dx = (— iy +1 J Q <fij tq (x)dx. It follows that for q even, i.e., q — q^, <fij iq (x) is even (odd) if j is odd (even). Hence for j and q both even, J Q <f>j^ q (x)dx = 0. Thus, as £„(s) G E, we have 1 (t) jtin ( s )(x)dx = Vje{2fc|fc>l}. '0 Therefore for n even, the heat content Q^ n ( s )(t) is given by o«-(.)(*) = E e ~^ 2t f / kwW* 1 ) • i odd VJ ° ' Given the fact that the flow £„(s) leaves any Dirichlet eigenvalue Xj(s) with j 7^ n invariant one could be tempted to believe that (J Q cf>j^ n ( S )(x)dx) 2 is indepen- dent of s. However, this is not the case. Using the formula for 2/2(0;, j 2 ir 2 , £ ra (s)) given in [181 p. Ill] one can prove that for j odd In fact, by a tedious but straightforward computation, one sees that for n even 2,2, 1/1 1 s.L=<A»m(*) = 8 E e 7 2 7T 2 V (7 2 — n 2 )7r 2 n 2 7r 2 jodd J vv ; 15 [2 implying that d s I ._ n Q£ re ( s ) 7^ 0. A similar conclusion holds in the case where n is odd. However, the additional summand j = n is more complicated, leading to the formula jodd, j^n xr 2 2 2 2/ --^e-" f + 8e-"' / d s | s _ o2/2 (z,ri 2 7r 2 + S ,£„( S ))(fe, where the last term can be explicitly computed using [T8l p. 111]. References !.] M. van den Berg, P. Gilkey Heat content asymptotics of a Riemannian manifold with boundary, J. Funct. Anal. 120, 48-71 (1994). M. van den Berg, P. Gilkey, K. Kirsten, V. A. Kozlov, Heat content asymp- totics for Riemannian manifolds with Zaremba boundary conditions, Poten- tial Analysis 26, 225-254 (2007). M. van den Berg, S. Srisatkunarajah, Heat equation for a region in K 2 with a polygonal boundary, J. Lond. Math. Soc. (2) 37 119-127 (1988). M. van den Berg, S. Srisatkunarajah, Heat flow and Brownian motion for a region in M. 2 with a polygonal boundary, Probab. Theor. Rel. Fields 86, 41-52 (1990). M. van den Berg, J.- F. Le Gall, Mean curvature and the heat equation, Math. Z. 215, 437-464 (1994). P. Buser, J. Conway, P. Doyle, K.-D. Semmler, Some planar isospectral domains, Intern. Math. Res. Notices 9 391-400 (1994). S. J. Chapman, Drums that sound the same, Amer. Math. Monthly 102, 124-138 (1995). P. Gilkey, Asymptotic Formulae in Spectral Geometry, Stud. Adv. Math., Chapmanfc Hall/CRC, Boca Raton, FL (2004). P. Gilkey, Heat content, heat trace and isospectrality, New developments in Lie theory and geometry, Contemp. Math. 491, 115-123 (2009). C. Gordon, P. Perry, D. Schueth, Isospectral and isoscattering manifolds: a survey of techniques and examples, Contemp. Math. 387, 157-179 (2005). C. Gordon, D. Webb, S. Wolpert, Isospectral plane domains via Rieman- nian orbifolds, Invent. Math. 110, 1-22 (1992). P. Hcrbrich, On inaudible properties of broken drums - Isospectral domains with mixed boundary conditions, http://arXiv.org/abs/llll.6789v2 (2012). D. Jakobson, M. Levitin, N. Nadirashvili, I. Poltcrovich, Spectral problems with mixed Dirichlet- Neumann boundary conditions: isospectrality and be- yond, J. Comput. Appl. Math. 194, 141-155 (2006). [4 [5] [6] [7 [9 [10 [11 [12 [13 16 [14] P. McDonald, R. Meyers, Isospectral polygons, planar graphs and heat con- tent, Proc. Amer. Math. Soc. 131, 3589-3599 (2003). [15] H. P. McKean, I. M. Singer, Curvature and the eigenvalues of the Laplacian, J. Diff. Geom. 1, 43-69 (1967). [16] M. Levitin, D. Vassiliev, Spectral asymptotics, renewal theorem, and the Berry conjecture for a class of fractals, Proc. Lond. Math. Soc. 72, 188- 214 (1996). [17] O. Parzanchevski, R. Band, Linear representations and isospectrality with boundary conditions, J. Geom. Anal. 20, 439-471 (2010). [18] J. Poschel, E. Trubowitz, Inverse spectral theory, Pure and Applied Math- ematics 130, Academic Press, Inc., Boston, MA, (1987). 17